xref: /linux/kernel/trace/trace_fprobe.c (revision b7dbc2e813e00d61e66fc0267599441493774b93)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Fprobe-based tracing events
4  * Copyright (C) 2022 Google LLC.
5  */
6 #define pr_fmt(fmt)	"trace_fprobe: " fmt
7 
8 #include <linux/fprobe.h>
9 #include <linux/list.h>
10 #include <linux/module.h>
11 #include <linux/mutex.h>
12 #include <linux/rculist.h>
13 #include <linux/security.h>
14 #include <linux/tracepoint.h>
15 #include <linux/uaccess.h>
16 
17 #include <asm/ptrace.h>
18 
19 #include "trace_dynevent.h"
20 #include "trace_probe.h"
21 #include "trace_probe_kernel.h"
22 #include "trace_probe_tmpl.h"
23 
24 #define FPROBE_EVENT_SYSTEM "fprobes"
25 #define TRACEPOINT_EVENT_SYSTEM "tracepoints"
26 #define RETHOOK_MAXACTIVE_MAX 4096
27 
28 static int trace_fprobe_create(const char *raw_command);
29 static int trace_fprobe_show(struct seq_file *m, struct dyn_event *ev);
30 static int trace_fprobe_release(struct dyn_event *ev);
31 static bool trace_fprobe_is_busy(struct dyn_event *ev);
32 static bool trace_fprobe_match(const char *system, const char *event,
33 			int argc, const char **argv, struct dyn_event *ev);
34 
35 static struct dyn_event_operations trace_fprobe_ops = {
36 	.create = trace_fprobe_create,
37 	.show = trace_fprobe_show,
38 	.is_busy = trace_fprobe_is_busy,
39 	.free = trace_fprobe_release,
40 	.match = trace_fprobe_match,
41 };
42 
43 /* List of tracepoint_user */
44 static LIST_HEAD(tracepoint_user_list);
45 static DEFINE_MUTEX(tracepoint_user_mutex);
46 
47 /* While living tracepoint_user, @tpoint can be NULL and @refcount != 0. */
48 struct tracepoint_user {
49 	struct list_head	list;
50 	const char		*name;
51 	struct tracepoint	*tpoint;
52 	unsigned int		refcount;
53 };
54 
55 /* NOTE: you must lock tracepoint_user_mutex. */
56 #define for_each_tracepoint_user(tuser)		\
57 	list_for_each_entry(tuser, &tracepoint_user_list, list)
58 
tracepoint_user_register(struct tracepoint_user * tuser)59 static int tracepoint_user_register(struct tracepoint_user *tuser)
60 {
61 	struct tracepoint *tpoint = tuser->tpoint;
62 
63 	if (!tpoint)
64 		return 0;
65 
66 	return tracepoint_probe_register_prio_may_exist(tpoint,
67 					tpoint->probestub, NULL, 0);
68 }
69 
tracepoint_user_unregister(struct tracepoint_user * tuser)70 static void tracepoint_user_unregister(struct tracepoint_user *tuser)
71 {
72 	if (!tuser->tpoint)
73 		return;
74 
75 	WARN_ON_ONCE(tracepoint_probe_unregister(tuser->tpoint, tuser->tpoint->probestub, NULL));
76 	tuser->tpoint = NULL;
77 }
78 
tracepoint_user_ip(struct tracepoint_user * tuser)79 static unsigned long tracepoint_user_ip(struct tracepoint_user *tuser)
80 {
81 	if (!tuser->tpoint)
82 		return 0UL;
83 
84 	return (unsigned long)tuser->tpoint->probestub;
85 }
86 
__tracepoint_user_free(struct tracepoint_user * tuser)87 static void __tracepoint_user_free(struct tracepoint_user *tuser)
88 {
89 	if (!tuser)
90 		return;
91 	kfree(tuser->name);
92 	kfree(tuser);
93 }
94 
DEFINE_FREE(tuser_free,struct tracepoint_user *,__tracepoint_user_free (_T))95 DEFINE_FREE(tuser_free, struct tracepoint_user *, __tracepoint_user_free(_T))
96 
97 static struct tracepoint_user *__tracepoint_user_init(const char *name, struct tracepoint *tpoint)
98 {
99 	struct tracepoint_user *tuser __free(tuser_free) = NULL;
100 	int ret;
101 
102 	tuser = kzalloc(sizeof(*tuser), GFP_KERNEL);
103 	if (!tuser)
104 		return NULL;
105 	tuser->name = kstrdup(name, GFP_KERNEL);
106 	if (!tuser->name)
107 		return NULL;
108 
109 	if (tpoint) {
110 		ret = tracepoint_user_register(tuser);
111 		if (ret)
112 			return ERR_PTR(ret);
113 	}
114 
115 	tuser->tpoint = tpoint;
116 	tuser->refcount = 1;
117 	INIT_LIST_HEAD(&tuser->list);
118 	list_add(&tuser->list, &tracepoint_user_list);
119 
120 	return_ptr(tuser);
121 }
122 
123 static struct tracepoint *find_tracepoint(const char *tp_name,
124 	struct module **tp_mod);
125 
126 /*
127  * Get tracepoint_user if exist, or allocate new one and register it.
128  * If tracepoint is on a module, get its refcounter too.
129  * This returns errno or NULL (not loaded yet) or tracepoint_user.
130  */
tracepoint_user_find_get(const char * name,struct module ** pmod)131 static struct tracepoint_user *tracepoint_user_find_get(const char *name, struct module **pmod)
132 {
133 	struct module *mod __free(module_put) = NULL;
134 	struct tracepoint_user *tuser;
135 	struct tracepoint *tpoint;
136 
137 	if (!name || !pmod)
138 		return ERR_PTR(-EINVAL);
139 
140 	/* Get and lock the module which has tracepoint. */
141 	tpoint = find_tracepoint(name, &mod);
142 
143 	guard(mutex)(&tracepoint_user_mutex);
144 	/* Search existing tracepoint_user */
145 	for_each_tracepoint_user(tuser) {
146 		if (!strcmp(tuser->name, name)) {
147 			tuser->refcount++;
148 			*pmod = no_free_ptr(mod);
149 			return tuser;
150 		}
151 	}
152 
153 	/* The corresponding tracepoint_user is not found. */
154 	tuser = __tracepoint_user_init(name, tpoint);
155 	if (!IS_ERR_OR_NULL(tuser))
156 		*pmod = no_free_ptr(mod);
157 
158 	return tuser;
159 }
160 
tracepoint_user_put(struct tracepoint_user * tuser)161 static void tracepoint_user_put(struct tracepoint_user *tuser)
162 {
163 	scoped_guard(mutex, &tracepoint_user_mutex) {
164 		if (--tuser->refcount > 0)
165 			return;
166 
167 		list_del(&tuser->list);
168 		tracepoint_user_unregister(tuser);
169 	}
170 
171 	__tracepoint_user_free(tuser);
172 }
173 
174 DEFINE_FREE(tuser_put, struct tracepoint_user *,
175 	if (!IS_ERR_OR_NULL(_T))
176 		tracepoint_user_put(_T))
177 
178 /*
179  * Fprobe event core functions
180  */
181 
182 /*
183  * @tprobe is true for tracepoint probe.
184  * @tuser can be NULL if the trace_fprobe is disabled or the tracepoint is not
185  * loaded with a module. If @tuser != NULL, this trace_fprobe is enabled.
186  */
187 struct trace_fprobe {
188 	struct dyn_event	devent;
189 	struct fprobe		fp;
190 	const char		*symbol;
191 	bool			tprobe;
192 	struct tracepoint_user	*tuser;
193 	struct trace_probe	tp;
194 };
195 
is_trace_fprobe(struct dyn_event * ev)196 static bool is_trace_fprobe(struct dyn_event *ev)
197 {
198 	return ev->ops == &trace_fprobe_ops;
199 }
200 
to_trace_fprobe(struct dyn_event * ev)201 static struct trace_fprobe *to_trace_fprobe(struct dyn_event *ev)
202 {
203 	return container_of(ev, struct trace_fprobe, devent);
204 }
205 
206 /**
207  * for_each_trace_fprobe - iterate over the trace_fprobe list
208  * @pos:	the struct trace_fprobe * for each entry
209  * @dpos:	the struct dyn_event * to use as a loop cursor
210  */
211 #define for_each_trace_fprobe(pos, dpos)	\
212 	for_each_dyn_event(dpos)		\
213 		if (is_trace_fprobe(dpos) && (pos = to_trace_fprobe(dpos)))
214 
trace_fprobe_is_return(struct trace_fprobe * tf)215 static bool trace_fprobe_is_return(struct trace_fprobe *tf)
216 {
217 	return tf->fp.exit_handler != NULL;
218 }
219 
trace_fprobe_is_tracepoint(struct trace_fprobe * tf)220 static bool trace_fprobe_is_tracepoint(struct trace_fprobe *tf)
221 {
222 	return tf->tprobe;
223 }
224 
trace_fprobe_symbol(struct trace_fprobe * tf)225 static const char *trace_fprobe_symbol(struct trace_fprobe *tf)
226 {
227 	return tf->symbol ? tf->symbol : "unknown";
228 }
229 
trace_fprobe_is_busy(struct dyn_event * ev)230 static bool trace_fprobe_is_busy(struct dyn_event *ev)
231 {
232 	struct trace_fprobe *tf = to_trace_fprobe(ev);
233 
234 	return trace_probe_is_enabled(&tf->tp);
235 }
236 
trace_fprobe_match_command_head(struct trace_fprobe * tf,int argc,const char ** argv)237 static bool trace_fprobe_match_command_head(struct trace_fprobe *tf,
238 					    int argc, const char **argv)
239 {
240 	char buf[MAX_ARGSTR_LEN + 1];
241 
242 	if (!argc)
243 		return true;
244 
245 	snprintf(buf, sizeof(buf), "%s", trace_fprobe_symbol(tf));
246 	if (strcmp(buf, argv[0]))
247 		return false;
248 	argc--; argv++;
249 
250 	return trace_probe_match_command_args(&tf->tp, argc, argv);
251 }
252 
trace_fprobe_match(const char * system,const char * event,int argc,const char ** argv,struct dyn_event * ev)253 static bool trace_fprobe_match(const char *system, const char *event,
254 			int argc, const char **argv, struct dyn_event *ev)
255 {
256 	struct trace_fprobe *tf = to_trace_fprobe(ev);
257 
258 	if (event[0] != '\0' && strcmp(trace_probe_name(&tf->tp), event))
259 		return false;
260 
261 	if (system && strcmp(trace_probe_group_name(&tf->tp), system))
262 		return false;
263 
264 	return trace_fprobe_match_command_head(tf, argc, argv);
265 }
266 
trace_fprobe_is_registered(struct trace_fprobe * tf)267 static bool trace_fprobe_is_registered(struct trace_fprobe *tf)
268 {
269 	return fprobe_is_registered(&tf->fp);
270 }
271 
272 /*
273  * Note that we don't verify the fetch_insn code, since it does not come
274  * from user space.
275  */
276 static int
process_fetch_insn(struct fetch_insn * code,void * rec,void * edata,void * dest,void * base)277 process_fetch_insn(struct fetch_insn *code, void *rec, void *edata,
278 		   void *dest, void *base)
279 {
280 	struct ftrace_regs *fregs = rec;
281 	unsigned long val;
282 	int ret;
283 
284 retry:
285 	/* 1st stage: get value from context */
286 	switch (code->op) {
287 	case FETCH_OP_STACK:
288 		val = ftrace_regs_get_kernel_stack_nth(fregs, code->param);
289 		break;
290 	case FETCH_OP_STACKP:
291 		val = ftrace_regs_get_stack_pointer(fregs);
292 		break;
293 	case FETCH_OP_RETVAL:
294 		val = ftrace_regs_get_return_value(fregs);
295 		break;
296 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
297 	case FETCH_OP_ARG:
298 		val = ftrace_regs_get_argument(fregs, code->param);
299 		break;
300 	case FETCH_OP_EDATA:
301 		val = *(unsigned long *)((unsigned long)edata + code->offset);
302 		break;
303 #endif
304 	case FETCH_NOP_SYMBOL:	/* Ignore a place holder */
305 		code++;
306 		goto retry;
307 	default:
308 		ret = process_common_fetch_insn(code, &val);
309 		if (ret < 0)
310 			return ret;
311 	}
312 	code++;
313 
314 	return process_fetch_insn_bottom(code, val, dest, base);
315 }
NOKPROBE_SYMBOL(process_fetch_insn)316 NOKPROBE_SYMBOL(process_fetch_insn)
317 
318 /* function entry handler */
319 static nokprobe_inline void
320 __fentry_trace_func(struct trace_fprobe *tf, unsigned long entry_ip,
321 		    struct ftrace_regs *fregs,
322 		    struct trace_event_file *trace_file)
323 {
324 	struct fentry_trace_entry_head *entry;
325 	struct trace_event_call *call = trace_probe_event_call(&tf->tp);
326 	struct trace_event_buffer fbuffer;
327 	int dsize;
328 
329 	if (WARN_ON_ONCE(call != trace_file->event_call))
330 		return;
331 
332 	if (trace_trigger_soft_disabled(trace_file))
333 		return;
334 
335 	dsize = __get_data_size(&tf->tp, fregs, NULL);
336 
337 	entry = trace_event_buffer_reserve(&fbuffer, trace_file,
338 					   sizeof(*entry) + tf->tp.size + dsize);
339 	if (!entry)
340 		return;
341 
342 	fbuffer.regs = ftrace_get_regs(fregs);
343 	entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
344 	entry->ip = entry_ip;
345 	store_trace_args(&entry[1], &tf->tp, fregs, NULL, sizeof(*entry), dsize);
346 
347 	trace_event_buffer_commit(&fbuffer);
348 }
349 
350 static void
fentry_trace_func(struct trace_fprobe * tf,unsigned long entry_ip,struct ftrace_regs * fregs)351 fentry_trace_func(struct trace_fprobe *tf, unsigned long entry_ip,
352 		  struct ftrace_regs *fregs)
353 {
354 	struct event_file_link *link;
355 
356 	trace_probe_for_each_link_rcu(link, &tf->tp)
357 		__fentry_trace_func(tf, entry_ip, fregs, link->file);
358 }
359 NOKPROBE_SYMBOL(fentry_trace_func);
360 
361 static nokprobe_inline
store_fprobe_entry_data(void * edata,struct trace_probe * tp,struct ftrace_regs * fregs)362 void store_fprobe_entry_data(void *edata, struct trace_probe *tp, struct ftrace_regs *fregs)
363 {
364 	struct probe_entry_arg *earg = tp->entry_arg;
365 	unsigned long val = 0;
366 	int i;
367 
368 	if (!earg)
369 		return;
370 
371 	for (i = 0; i < earg->size; i++) {
372 		struct fetch_insn *code = &earg->code[i];
373 
374 		switch (code->op) {
375 		case FETCH_OP_ARG:
376 			val = ftrace_regs_get_argument(fregs, code->param);
377 			break;
378 		case FETCH_OP_ST_EDATA:
379 			*(unsigned long *)((unsigned long)edata + code->offset) = val;
380 			break;
381 		case FETCH_OP_END:
382 			goto end;
383 		default:
384 			break;
385 		}
386 	}
387 end:
388 	return;
389 }
390 
391 /* function exit handler */
trace_fprobe_entry_handler(struct fprobe * fp,unsigned long entry_ip,unsigned long ret_ip,struct ftrace_regs * fregs,void * entry_data)392 static int trace_fprobe_entry_handler(struct fprobe *fp, unsigned long entry_ip,
393 				unsigned long ret_ip, struct ftrace_regs *fregs,
394 				void *entry_data)
395 {
396 	struct trace_fprobe *tf = container_of(fp, struct trace_fprobe, fp);
397 
398 	if (tf->tp.entry_arg)
399 		store_fprobe_entry_data(entry_data, &tf->tp, fregs);
400 
401 	return 0;
402 }
NOKPROBE_SYMBOL(trace_fprobe_entry_handler)403 NOKPROBE_SYMBOL(trace_fprobe_entry_handler)
404 
405 static nokprobe_inline void
406 __fexit_trace_func(struct trace_fprobe *tf, unsigned long entry_ip,
407 		   unsigned long ret_ip, struct ftrace_regs *fregs,
408 		   void *entry_data, struct trace_event_file *trace_file)
409 {
410 	struct fexit_trace_entry_head *entry;
411 	struct trace_event_buffer fbuffer;
412 	struct trace_event_call *call = trace_probe_event_call(&tf->tp);
413 	int dsize;
414 
415 	if (WARN_ON_ONCE(call != trace_file->event_call))
416 		return;
417 
418 	if (trace_trigger_soft_disabled(trace_file))
419 		return;
420 
421 	dsize = __get_data_size(&tf->tp, fregs, entry_data);
422 
423 	entry = trace_event_buffer_reserve(&fbuffer, trace_file,
424 					   sizeof(*entry) + tf->tp.size + dsize);
425 	if (!entry)
426 		return;
427 
428 	fbuffer.regs = ftrace_get_regs(fregs);
429 	entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
430 	entry->func = entry_ip;
431 	entry->ret_ip = ret_ip;
432 	store_trace_args(&entry[1], &tf->tp, fregs, entry_data, sizeof(*entry), dsize);
433 
434 	trace_event_buffer_commit(&fbuffer);
435 }
436 
437 static void
fexit_trace_func(struct trace_fprobe * tf,unsigned long entry_ip,unsigned long ret_ip,struct ftrace_regs * fregs,void * entry_data)438 fexit_trace_func(struct trace_fprobe *tf, unsigned long entry_ip,
439 		 unsigned long ret_ip, struct ftrace_regs *fregs, void *entry_data)
440 {
441 	struct event_file_link *link;
442 
443 	trace_probe_for_each_link_rcu(link, &tf->tp)
444 		__fexit_trace_func(tf, entry_ip, ret_ip, fregs, entry_data, link->file);
445 }
446 NOKPROBE_SYMBOL(fexit_trace_func);
447 
448 #ifdef CONFIG_PERF_EVENTS
449 
fentry_perf_func(struct trace_fprobe * tf,unsigned long entry_ip,struct ftrace_regs * fregs)450 static int fentry_perf_func(struct trace_fprobe *tf, unsigned long entry_ip,
451 			    struct ftrace_regs *fregs)
452 {
453 	struct trace_event_call *call = trace_probe_event_call(&tf->tp);
454 	struct fentry_trace_entry_head *entry;
455 	struct hlist_head *head;
456 	int size, __size, dsize;
457 	struct pt_regs *regs;
458 	int rctx;
459 
460 	head = this_cpu_ptr(call->perf_events);
461 	if (hlist_empty(head))
462 		return 0;
463 
464 	dsize = __get_data_size(&tf->tp, fregs, NULL);
465 	__size = sizeof(*entry) + tf->tp.size + dsize;
466 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
467 	size -= sizeof(u32);
468 
469 	entry = perf_trace_buf_alloc(size, &regs, &rctx);
470 	if (!entry)
471 		return 0;
472 
473 	regs = ftrace_fill_perf_regs(fregs, regs);
474 
475 	entry->ip = entry_ip;
476 	memset(&entry[1], 0, dsize);
477 	store_trace_args(&entry[1], &tf->tp, fregs, NULL, sizeof(*entry), dsize);
478 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
479 			      head, NULL);
480 	return 0;
481 }
482 NOKPROBE_SYMBOL(fentry_perf_func);
483 
484 static void
fexit_perf_func(struct trace_fprobe * tf,unsigned long entry_ip,unsigned long ret_ip,struct ftrace_regs * fregs,void * entry_data)485 fexit_perf_func(struct trace_fprobe *tf, unsigned long entry_ip,
486 		unsigned long ret_ip, struct ftrace_regs *fregs,
487 		void *entry_data)
488 {
489 	struct trace_event_call *call = trace_probe_event_call(&tf->tp);
490 	struct fexit_trace_entry_head *entry;
491 	struct hlist_head *head;
492 	int size, __size, dsize;
493 	struct pt_regs *regs;
494 	int rctx;
495 
496 	head = this_cpu_ptr(call->perf_events);
497 	if (hlist_empty(head))
498 		return;
499 
500 	dsize = __get_data_size(&tf->tp, fregs, entry_data);
501 	__size = sizeof(*entry) + tf->tp.size + dsize;
502 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
503 	size -= sizeof(u32);
504 
505 	entry = perf_trace_buf_alloc(size, &regs, &rctx);
506 	if (!entry)
507 		return;
508 
509 	regs = ftrace_fill_perf_regs(fregs, regs);
510 
511 	entry->func = entry_ip;
512 	entry->ret_ip = ret_ip;
513 	store_trace_args(&entry[1], &tf->tp, fregs, entry_data, sizeof(*entry), dsize);
514 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
515 			      head, NULL);
516 }
517 NOKPROBE_SYMBOL(fexit_perf_func);
518 #endif	/* CONFIG_PERF_EVENTS */
519 
fentry_dispatcher(struct fprobe * fp,unsigned long entry_ip,unsigned long ret_ip,struct ftrace_regs * fregs,void * entry_data)520 static int fentry_dispatcher(struct fprobe *fp, unsigned long entry_ip,
521 			     unsigned long ret_ip, struct ftrace_regs *fregs,
522 			     void *entry_data)
523 {
524 	struct trace_fprobe *tf = container_of(fp, struct trace_fprobe, fp);
525 	int ret = 0;
526 
527 	if (trace_probe_test_flag(&tf->tp, TP_FLAG_TRACE))
528 		fentry_trace_func(tf, entry_ip, fregs);
529 
530 #ifdef CONFIG_PERF_EVENTS
531 	if (trace_probe_test_flag(&tf->tp, TP_FLAG_PROFILE))
532 		ret = fentry_perf_func(tf, entry_ip, fregs);
533 #endif
534 	return ret;
535 }
536 NOKPROBE_SYMBOL(fentry_dispatcher);
537 
fexit_dispatcher(struct fprobe * fp,unsigned long entry_ip,unsigned long ret_ip,struct ftrace_regs * fregs,void * entry_data)538 static void fexit_dispatcher(struct fprobe *fp, unsigned long entry_ip,
539 			     unsigned long ret_ip, struct ftrace_regs *fregs,
540 			     void *entry_data)
541 {
542 	struct trace_fprobe *tf = container_of(fp, struct trace_fprobe, fp);
543 
544 	if (trace_probe_test_flag(&tf->tp, TP_FLAG_TRACE))
545 		fexit_trace_func(tf, entry_ip, ret_ip, fregs, entry_data);
546 #ifdef CONFIG_PERF_EVENTS
547 	if (trace_probe_test_flag(&tf->tp, TP_FLAG_PROFILE))
548 		fexit_perf_func(tf, entry_ip, ret_ip, fregs, entry_data);
549 #endif
550 }
551 NOKPROBE_SYMBOL(fexit_dispatcher);
552 
free_trace_fprobe(struct trace_fprobe * tf)553 static void free_trace_fprobe(struct trace_fprobe *tf)
554 {
555 	if (tf) {
556 		trace_probe_cleanup(&tf->tp);
557 		if (tf->tuser)
558 			tracepoint_user_put(tf->tuser);
559 		kfree(tf->symbol);
560 		kfree(tf);
561 	}
562 }
563 
564 /* Since alloc_trace_fprobe() can return error, check the pointer is ERR too. */
565 DEFINE_FREE(free_trace_fprobe, struct trace_fprobe *, if (!IS_ERR_OR_NULL(_T)) free_trace_fprobe(_T))
566 
567 /*
568  * Allocate new trace_probe and initialize it (including fprobe).
569  */
alloc_trace_fprobe(const char * group,const char * event,const char * symbol,int nargs,bool is_return,bool is_tracepoint)570 static struct trace_fprobe *alloc_trace_fprobe(const char *group,
571 					       const char *event,
572 					       const char *symbol,
573 					       int nargs, bool is_return,
574 					       bool is_tracepoint)
575 {
576 	struct trace_fprobe *tf __free(free_trace_fprobe) = NULL;
577 	int ret = -ENOMEM;
578 
579 	tf = kzalloc(struct_size(tf, tp.args, nargs), GFP_KERNEL);
580 	if (!tf)
581 		return ERR_PTR(ret);
582 
583 	tf->symbol = kstrdup(symbol, GFP_KERNEL);
584 	if (!tf->symbol)
585 		return ERR_PTR(-ENOMEM);
586 
587 	if (is_return)
588 		tf->fp.exit_handler = fexit_dispatcher;
589 	else
590 		tf->fp.entry_handler = fentry_dispatcher;
591 
592 	tf->tprobe = is_tracepoint;
593 
594 	ret = trace_probe_init(&tf->tp, event, group, false, nargs);
595 	if (ret < 0)
596 		return ERR_PTR(ret);
597 
598 	dyn_event_init(&tf->devent, &trace_fprobe_ops);
599 	return_ptr(tf);
600 }
601 
find_trace_fprobe(const char * event,const char * group)602 static struct trace_fprobe *find_trace_fprobe(const char *event,
603 					      const char *group)
604 {
605 	struct dyn_event *pos;
606 	struct trace_fprobe *tf;
607 
608 	for_each_trace_fprobe(tf, pos)
609 		if (strcmp(trace_probe_name(&tf->tp), event) == 0 &&
610 		    strcmp(trace_probe_group_name(&tf->tp), group) == 0)
611 			return tf;
612 	return NULL;
613 }
614 
615 /* Event entry printers */
616 static enum print_line_t
print_fentry_event(struct trace_iterator * iter,int flags,struct trace_event * event)617 print_fentry_event(struct trace_iterator *iter, int flags,
618 		   struct trace_event *event)
619 {
620 	struct fentry_trace_entry_head *field;
621 	struct trace_seq *s = &iter->seq;
622 	struct trace_probe *tp;
623 
624 	field = (struct fentry_trace_entry_head *)iter->ent;
625 	tp = trace_probe_primary_from_call(
626 		container_of(event, struct trace_event_call, event));
627 	if (WARN_ON_ONCE(!tp))
628 		goto out;
629 
630 	trace_seq_printf(s, "%s: (", trace_probe_name(tp));
631 
632 	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
633 		goto out;
634 
635 	trace_seq_putc(s, ')');
636 
637 	if (trace_probe_print_args(s, tp->args, tp->nr_args,
638 			     (u8 *)&field[1], field) < 0)
639 		goto out;
640 
641 	trace_seq_putc(s, '\n');
642  out:
643 	return trace_handle_return(s);
644 }
645 
646 static enum print_line_t
print_fexit_event(struct trace_iterator * iter,int flags,struct trace_event * event)647 print_fexit_event(struct trace_iterator *iter, int flags,
648 		  struct trace_event *event)
649 {
650 	struct fexit_trace_entry_head *field;
651 	struct trace_seq *s = &iter->seq;
652 	struct trace_probe *tp;
653 
654 	field = (struct fexit_trace_entry_head *)iter->ent;
655 	tp = trace_probe_primary_from_call(
656 		container_of(event, struct trace_event_call, event));
657 	if (WARN_ON_ONCE(!tp))
658 		goto out;
659 
660 	trace_seq_printf(s, "%s: (", trace_probe_name(tp));
661 
662 	if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
663 		goto out;
664 
665 	trace_seq_puts(s, " <- ");
666 
667 	if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
668 		goto out;
669 
670 	trace_seq_putc(s, ')');
671 
672 	if (trace_probe_print_args(s, tp->args, tp->nr_args,
673 			     (u8 *)&field[1], field) < 0)
674 		goto out;
675 
676 	trace_seq_putc(s, '\n');
677 
678  out:
679 	return trace_handle_return(s);
680 }
681 
fentry_event_define_fields(struct trace_event_call * event_call)682 static int fentry_event_define_fields(struct trace_event_call *event_call)
683 {
684 	int ret;
685 	struct fentry_trace_entry_head field;
686 	struct trace_probe *tp;
687 
688 	tp = trace_probe_primary_from_call(event_call);
689 	if (WARN_ON_ONCE(!tp))
690 		return -ENOENT;
691 
692 	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
693 
694 	return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
695 }
696 
fexit_event_define_fields(struct trace_event_call * event_call)697 static int fexit_event_define_fields(struct trace_event_call *event_call)
698 {
699 	int ret;
700 	struct fexit_trace_entry_head field;
701 	struct trace_probe *tp;
702 
703 	tp = trace_probe_primary_from_call(event_call);
704 	if (WARN_ON_ONCE(!tp))
705 		return -ENOENT;
706 
707 	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
708 	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
709 
710 	return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
711 }
712 
713 static struct trace_event_functions fentry_funcs = {
714 	.trace		= print_fentry_event
715 };
716 
717 static struct trace_event_functions fexit_funcs = {
718 	.trace		= print_fexit_event
719 };
720 
721 static struct trace_event_fields fentry_fields_array[] = {
722 	{ .type = TRACE_FUNCTION_TYPE,
723 	  .define_fields = fentry_event_define_fields },
724 	{}
725 };
726 
727 static struct trace_event_fields fexit_fields_array[] = {
728 	{ .type = TRACE_FUNCTION_TYPE,
729 	  .define_fields = fexit_event_define_fields },
730 	{}
731 };
732 
733 static int fprobe_register(struct trace_event_call *event,
734 			   enum trace_reg type, void *data);
735 
init_trace_event_call(struct trace_fprobe * tf)736 static inline void init_trace_event_call(struct trace_fprobe *tf)
737 {
738 	struct trace_event_call *call = trace_probe_event_call(&tf->tp);
739 
740 	if (trace_fprobe_is_return(tf)) {
741 		call->event.funcs = &fexit_funcs;
742 		call->class->fields_array = fexit_fields_array;
743 	} else {
744 		call->event.funcs = &fentry_funcs;
745 		call->class->fields_array = fentry_fields_array;
746 	}
747 
748 	call->flags = TRACE_EVENT_FL_FPROBE;
749 	call->class->reg = fprobe_register;
750 }
751 
register_fprobe_event(struct trace_fprobe * tf)752 static int register_fprobe_event(struct trace_fprobe *tf)
753 {
754 	init_trace_event_call(tf);
755 
756 	return trace_probe_register_event_call(&tf->tp);
757 }
758 
unregister_fprobe_event(struct trace_fprobe * tf)759 static int unregister_fprobe_event(struct trace_fprobe *tf)
760 {
761 	return trace_probe_unregister_event_call(&tf->tp);
762 }
763 
__regsiter_tracepoint_fprobe(struct trace_fprobe * tf)764 static int __regsiter_tracepoint_fprobe(struct trace_fprobe *tf)
765 {
766 	struct tracepoint_user *tuser __free(tuser_put) = NULL;
767 	struct module *mod __free(module_put) = NULL;
768 	unsigned long ip;
769 	int ret;
770 
771 	if (WARN_ON_ONCE(tf->tuser))
772 		return -EINVAL;
773 
774 	/* If the tracepoint is in a module, it must be locked in this function. */
775 	tuser = tracepoint_user_find_get(tf->symbol, &mod);
776 	/* This tracepoint is not loaded yet */
777 	if (IS_ERR(tuser))
778 		return PTR_ERR(tuser);
779 	if (!tuser)
780 		return -ENOMEM;
781 
782 	/* Register fprobe only if the tracepoint is loaded. */
783 	if (tuser->tpoint) {
784 		ip = tracepoint_user_ip(tuser);
785 		if (WARN_ON_ONCE(!ip))
786 			return -ENOENT;
787 
788 		ret = register_fprobe_ips(&tf->fp, &ip, 1);
789 		if (ret < 0)
790 			return ret;
791 	}
792 
793 	tf->tuser = no_free_ptr(tuser);
794 	return 0;
795 }
796 
797 /* Returns an error if the target function is not available, or 0 */
trace_fprobe_verify_target(struct trace_fprobe * tf)798 static int trace_fprobe_verify_target(struct trace_fprobe *tf)
799 {
800 	int ret;
801 
802 	/* Tracepoint should have a stub function. */
803 	if (trace_fprobe_is_tracepoint(tf))
804 		return 0;
805 
806 	/*
807 	 * Note: since we don't lock the module, even if this succeeded,
808 	 * register_fprobe() later can fail.
809 	 */
810 	ret = fprobe_count_ips_from_filter(tf->symbol, NULL);
811 	return (ret < 0) ? ret : 0;
812 }
813 
814 /* Internal register function - just handle fprobe and flags */
__register_trace_fprobe(struct trace_fprobe * tf)815 static int __register_trace_fprobe(struct trace_fprobe *tf)
816 {
817 	int i, ret;
818 
819 	/* Should we need new LOCKDOWN flag for fprobe? */
820 	ret = security_locked_down(LOCKDOWN_KPROBES);
821 	if (ret)
822 		return ret;
823 
824 	if (trace_fprobe_is_registered(tf))
825 		return -EINVAL;
826 
827 	for (i = 0; i < tf->tp.nr_args; i++) {
828 		ret = traceprobe_update_arg(&tf->tp.args[i]);
829 		if (ret)
830 			return ret;
831 	}
832 
833 	tf->fp.flags &= ~FPROBE_FL_DISABLED;
834 
835 	if (trace_fprobe_is_tracepoint(tf))
836 		return __regsiter_tracepoint_fprobe(tf);
837 
838 	/* TODO: handle filter, nofilter or symbol list */
839 	return register_fprobe(&tf->fp, tf->symbol, NULL);
840 }
841 
842 /* Internal unregister function - just handle fprobe and flags */
__unregister_trace_fprobe(struct trace_fprobe * tf)843 static void __unregister_trace_fprobe(struct trace_fprobe *tf)
844 {
845 	if (trace_fprobe_is_registered(tf))
846 		unregister_fprobe(&tf->fp);
847 	if (tf->tuser) {
848 		tracepoint_user_put(tf->tuser);
849 		tf->tuser = NULL;
850 	}
851 }
852 
853 /* TODO: make this trace_*probe common function */
854 /* Unregister a trace_probe and probe_event */
unregister_trace_fprobe(struct trace_fprobe * tf)855 static int unregister_trace_fprobe(struct trace_fprobe *tf)
856 {
857 	/* If other probes are on the event, just unregister fprobe */
858 	if (trace_probe_has_sibling(&tf->tp))
859 		goto unreg;
860 
861 	/* Enabled event can not be unregistered */
862 	if (trace_probe_is_enabled(&tf->tp))
863 		return -EBUSY;
864 
865 	/* If there's a reference to the dynamic event */
866 	if (trace_event_dyn_busy(trace_probe_event_call(&tf->tp)))
867 		return -EBUSY;
868 
869 	/* Will fail if probe is being used by ftrace or perf */
870 	if (unregister_fprobe_event(tf))
871 		return -EBUSY;
872 
873 unreg:
874 	__unregister_trace_fprobe(tf);
875 	dyn_event_remove(&tf->devent);
876 	trace_probe_unlink(&tf->tp);
877 
878 	return 0;
879 }
880 
trace_fprobe_has_same_fprobe(struct trace_fprobe * orig,struct trace_fprobe * comp)881 static bool trace_fprobe_has_same_fprobe(struct trace_fprobe *orig,
882 					 struct trace_fprobe *comp)
883 {
884 	struct trace_probe_event *tpe = orig->tp.event;
885 	int i;
886 
887 	list_for_each_entry(orig, &tpe->probes, tp.list) {
888 		if (strcmp(trace_fprobe_symbol(orig),
889 			   trace_fprobe_symbol(comp)))
890 			continue;
891 
892 		/*
893 		 * trace_probe_compare_arg_type() ensured that nr_args and
894 		 * each argument name and type are same. Let's compare comm.
895 		 */
896 		for (i = 0; i < orig->tp.nr_args; i++) {
897 			if (strcmp(orig->tp.args[i].comm,
898 				   comp->tp.args[i].comm))
899 				break;
900 		}
901 
902 		if (i == orig->tp.nr_args)
903 			return true;
904 	}
905 
906 	return false;
907 }
908 
append_trace_fprobe_event(struct trace_fprobe * tf,struct trace_fprobe * to)909 static int append_trace_fprobe_event(struct trace_fprobe *tf, struct trace_fprobe *to)
910 {
911 	int ret;
912 
913 	if (trace_fprobe_is_return(tf) != trace_fprobe_is_return(to) ||
914 	    trace_fprobe_is_tracepoint(tf) != trace_fprobe_is_tracepoint(to)) {
915 		trace_probe_log_set_index(0);
916 		trace_probe_log_err(0, DIFF_PROBE_TYPE);
917 		return -EEXIST;
918 	}
919 	ret = trace_probe_compare_arg_type(&tf->tp, &to->tp);
920 	if (ret) {
921 		/* Note that argument starts index = 2 */
922 		trace_probe_log_set_index(ret + 1);
923 		trace_probe_log_err(0, DIFF_ARG_TYPE);
924 		return -EEXIST;
925 	}
926 	if (trace_fprobe_has_same_fprobe(to, tf)) {
927 		trace_probe_log_set_index(0);
928 		trace_probe_log_err(0, SAME_PROBE);
929 		return -EEXIST;
930 	}
931 
932 	/* Append to existing event */
933 	ret = trace_probe_append(&tf->tp, &to->tp);
934 	if (ret)
935 		return ret;
936 
937 	ret = trace_fprobe_verify_target(tf);
938 	if (ret)
939 		trace_probe_unlink(&tf->tp);
940 	else
941 		dyn_event_add(&tf->devent, trace_probe_event_call(&tf->tp));
942 
943 	return ret;
944 }
945 
946 /* Register a trace_probe and probe_event, and check the fprobe is available. */
register_trace_fprobe_event(struct trace_fprobe * tf)947 static int register_trace_fprobe_event(struct trace_fprobe *tf)
948 {
949 	struct trace_fprobe *old_tf;
950 	int ret;
951 
952 	guard(mutex)(&event_mutex);
953 
954 	old_tf = find_trace_fprobe(trace_probe_name(&tf->tp),
955 				   trace_probe_group_name(&tf->tp));
956 	if (old_tf)
957 		return append_trace_fprobe_event(tf, old_tf);
958 
959 	/* Register new event */
960 	ret = register_fprobe_event(tf);
961 	if (ret) {
962 		if (ret == -EEXIST) {
963 			trace_probe_log_set_index(0);
964 			trace_probe_log_err(0, EVENT_EXIST);
965 		} else
966 			pr_warn("Failed to register probe event(%d)\n", ret);
967 		return ret;
968 	}
969 
970 	/* Verify fprobe is sane. */
971 	ret = trace_fprobe_verify_target(tf);
972 	if (ret < 0)
973 		unregister_fprobe_event(tf);
974 	else
975 		dyn_event_add(&tf->devent, trace_probe_event_call(&tf->tp));
976 
977 	return ret;
978 }
979 
980 struct __find_tracepoint_cb_data {
981 	const char *tp_name;
982 	struct tracepoint *tpoint;
983 	struct module *mod;
984 };
985 
__find_tracepoint_module_cb(struct tracepoint * tp,struct module * mod,void * priv)986 static void __find_tracepoint_module_cb(struct tracepoint *tp, struct module *mod, void *priv)
987 {
988 	struct __find_tracepoint_cb_data *data = priv;
989 
990 	if (!data->tpoint && !strcmp(data->tp_name, tp->name)) {
991 		/* If module is not specified, try getting module refcount. */
992 		if (!data->mod && mod) {
993 			/* If failed to get refcount, ignore this tracepoint. */
994 			if (!try_module_get(mod))
995 				return;
996 
997 			data->mod = mod;
998 		}
999 		data->tpoint = tp;
1000 	}
1001 }
1002 
__find_tracepoint_cb(struct tracepoint * tp,void * priv)1003 static void __find_tracepoint_cb(struct tracepoint *tp, void *priv)
1004 {
1005 	struct __find_tracepoint_cb_data *data = priv;
1006 
1007 	if (!data->tpoint && !strcmp(data->tp_name, tp->name))
1008 		data->tpoint = tp;
1009 }
1010 
1011 /*
1012  * Find a tracepoint from kernel and module. If the tracepoint is on the module,
1013  * the module's refcount is incremented and returned as *@tp_mod. Thus, if it is
1014  * not NULL, caller must call module_put(*tp_mod) after used the tracepoint.
1015  */
find_tracepoint(const char * tp_name,struct module ** tp_mod)1016 static struct tracepoint *find_tracepoint(const char *tp_name,
1017 					  struct module **tp_mod)
1018 {
1019 	struct __find_tracepoint_cb_data data = {
1020 		.tp_name = tp_name,
1021 		.mod = NULL,
1022 	};
1023 
1024 	for_each_kernel_tracepoint(__find_tracepoint_cb, &data);
1025 
1026 	if (!data.tpoint && IS_ENABLED(CONFIG_MODULES)) {
1027 		for_each_module_tracepoint(__find_tracepoint_module_cb, &data);
1028 		*tp_mod = data.mod;
1029 	}
1030 
1031 	return data.tpoint;
1032 }
1033 
1034 #ifdef CONFIG_MODULES
1035 /*
1036  * Find a tracepoint from specified module. In this case, this does not get the
1037  * module's refcount. The caller must ensure the module is not freed.
1038  */
find_tracepoint_in_module(struct module * mod,const char * tp_name)1039 static struct tracepoint *find_tracepoint_in_module(struct module *mod,
1040 						    const char *tp_name)
1041 {
1042 	struct __find_tracepoint_cb_data data = {
1043 		.tp_name = tp_name,
1044 		.mod = mod,
1045 	};
1046 
1047 	for_each_tracepoint_in_module(mod, __find_tracepoint_module_cb, &data);
1048 	return data.tpoint;
1049 }
1050 
1051 /* These are CONFIG_MODULES=y specific functions. */
tracepoint_user_within_module(struct tracepoint_user * tuser,struct module * mod)1052 static bool tracepoint_user_within_module(struct tracepoint_user *tuser,
1053 					  struct module *mod)
1054 {
1055 	return within_module(tracepoint_user_ip(tuser), mod);
1056 }
1057 
tracepoint_user_register_again(struct tracepoint_user * tuser,struct tracepoint * tpoint)1058 static int tracepoint_user_register_again(struct tracepoint_user *tuser,
1059 					  struct tracepoint *tpoint)
1060 {
1061 	tuser->tpoint = tpoint;
1062 	return tracepoint_user_register(tuser);
1063 }
1064 
tracepoint_user_unregister_clear(struct tracepoint_user * tuser)1065 static void tracepoint_user_unregister_clear(struct tracepoint_user *tuser)
1066 {
1067 	tracepoint_user_unregister(tuser);
1068 	tuser->tpoint = NULL;
1069 }
1070 
1071 /* module callback for tracepoint_user */
__tracepoint_probe_module_cb(struct notifier_block * self,unsigned long val,void * data)1072 static int __tracepoint_probe_module_cb(struct notifier_block *self,
1073 					unsigned long val, void *data)
1074 {
1075 	struct tp_module *tp_mod = data;
1076 	struct tracepoint_user *tuser;
1077 	struct tracepoint *tpoint;
1078 
1079 	if (val != MODULE_STATE_GOING && val != MODULE_STATE_COMING)
1080 		return NOTIFY_DONE;
1081 
1082 	mutex_lock(&tracepoint_user_mutex);
1083 	for_each_tracepoint_user(tuser) {
1084 		if (val == MODULE_STATE_COMING) {
1085 			/* This is not a tracepoint in this module. Skip it. */
1086 			tpoint = find_tracepoint_in_module(tp_mod->mod, tuser->name);
1087 			if (!tpoint)
1088 				continue;
1089 			WARN_ON_ONCE(tracepoint_user_register_again(tuser, tpoint));
1090 		} else if (val == MODULE_STATE_GOING &&
1091 			  tracepoint_user_within_module(tuser, tp_mod->mod)) {
1092 			/* Unregister all tracepoint_user in this module. */
1093 			tracepoint_user_unregister_clear(tuser);
1094 		}
1095 	}
1096 	mutex_unlock(&tracepoint_user_mutex);
1097 
1098 	return NOTIFY_DONE;
1099 }
1100 
1101 static struct notifier_block tracepoint_module_nb = {
1102 	.notifier_call = __tracepoint_probe_module_cb,
1103 };
1104 
1105 /* module callback for tprobe events */
__tprobe_event_module_cb(struct notifier_block * self,unsigned long val,void * data)1106 static int __tprobe_event_module_cb(struct notifier_block *self,
1107 				     unsigned long val, void *data)
1108 {
1109 	struct trace_fprobe *tf;
1110 	struct dyn_event *pos;
1111 	struct module *mod = data;
1112 
1113 	if (val != MODULE_STATE_GOING && val != MODULE_STATE_COMING)
1114 		return NOTIFY_DONE;
1115 
1116 	mutex_lock(&event_mutex);
1117 	for_each_trace_fprobe(tf, pos) {
1118 		/* Skip fprobe and disabled tprobe events. */
1119 		if (!trace_fprobe_is_tracepoint(tf) || !tf->tuser)
1120 			continue;
1121 
1122 		/* Before this notification, tracepoint notifier has already done. */
1123 		if (val == MODULE_STATE_COMING &&
1124 		    tracepoint_user_within_module(tf->tuser, mod)) {
1125 			unsigned long ip = tracepoint_user_ip(tf->tuser);
1126 
1127 			WARN_ON_ONCE(register_fprobe_ips(&tf->fp, &ip, 1));
1128 		} else if (val == MODULE_STATE_GOING &&
1129 			   /*
1130 			    * tracepoint_user_within_module() does not work here because
1131 			    * tracepoint_user is already unregistered and cleared tpoint.
1132 			    * Instead, checking whether the fprobe is registered but
1133 			    * tpoint is cleared(unregistered). Such unbalance probes
1134 			    * must be adjusted anyway.
1135 			    */
1136 			    trace_fprobe_is_registered(tf) &&
1137 			    !tf->tuser->tpoint) {
1138 			unregister_fprobe(&tf->fp);
1139 		}
1140 	}
1141 	mutex_unlock(&event_mutex);
1142 
1143 	return NOTIFY_DONE;
1144 }
1145 
1146 /* NOTE: this must be called after tracepoint callback */
1147 static struct notifier_block tprobe_event_module_nb = {
1148 	.notifier_call = __tprobe_event_module_cb,
1149 	/* Make sure this is later than tracepoint module notifier. */
1150 	.priority = -10,
1151 };
1152 #endif /* CONFIG_MODULES */
1153 
parse_symbol_and_return(int argc,const char * argv[],char ** symbol,bool * is_return,bool is_tracepoint)1154 static int parse_symbol_and_return(int argc, const char *argv[],
1155 				   char **symbol, bool *is_return,
1156 				   bool is_tracepoint)
1157 {
1158 	char *tmp = strchr(argv[1], '%');
1159 	int i;
1160 
1161 	if (tmp) {
1162 		int len = tmp - argv[1];
1163 
1164 		if (!is_tracepoint && !strcmp(tmp, "%return")) {
1165 			*is_return = true;
1166 		} else {
1167 			trace_probe_log_err(len, BAD_ADDR_SUFFIX);
1168 			return -EINVAL;
1169 		}
1170 		*symbol = kmemdup_nul(argv[1], len, GFP_KERNEL);
1171 	} else
1172 		*symbol = kstrdup(argv[1], GFP_KERNEL);
1173 	if (!*symbol)
1174 		return -ENOMEM;
1175 
1176 	if (*is_return)
1177 		return 0;
1178 
1179 	if (is_tracepoint) {
1180 		tmp = *symbol;
1181 		while (*tmp && (isalnum(*tmp) || *tmp == '_'))
1182 			tmp++;
1183 		if (*tmp) {
1184 			/* find a wrong character. */
1185 			trace_probe_log_err(tmp - *symbol, BAD_TP_NAME);
1186 			kfree(*symbol);
1187 			*symbol = NULL;
1188 			return -EINVAL;
1189 		}
1190 	}
1191 
1192 	/* If there is $retval, this should be a return fprobe. */
1193 	for (i = 2; i < argc; i++) {
1194 		tmp = strstr(argv[i], "$retval");
1195 		if (tmp && !isalnum(tmp[7]) && tmp[7] != '_') {
1196 			if (is_tracepoint) {
1197 				trace_probe_log_set_index(i);
1198 				trace_probe_log_err(tmp - argv[i], RETVAL_ON_PROBE);
1199 				kfree(*symbol);
1200 				*symbol = NULL;
1201 				return -EINVAL;
1202 			}
1203 			*is_return = true;
1204 			break;
1205 		}
1206 	}
1207 	return 0;
1208 }
1209 
trace_fprobe_create_internal(int argc,const char * argv[],struct traceprobe_parse_context * ctx)1210 static int trace_fprobe_create_internal(int argc, const char *argv[],
1211 					struct traceprobe_parse_context *ctx)
1212 {
1213 	/*
1214 	 * Argument syntax:
1215 	 *  - Add fentry probe:
1216 	 *      f[:[GRP/][EVENT]] [MOD:]KSYM [FETCHARGS]
1217 	 *  - Add fexit probe:
1218 	 *      f[N][:[GRP/][EVENT]] [MOD:]KSYM%return [FETCHARGS]
1219 	 *  - Add tracepoint probe:
1220 	 *      t[:[GRP/][EVENT]] TRACEPOINT [FETCHARGS]
1221 	 *
1222 	 * Fetch args:
1223 	 *  $retval	: fetch return value
1224 	 *  $stack	: fetch stack address
1225 	 *  $stackN	: fetch Nth entry of stack (N:0-)
1226 	 *  $argN	: fetch Nth argument (N:1-)
1227 	 *  $comm       : fetch current task comm
1228 	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
1229 	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
1230 	 * Dereferencing memory fetch:
1231 	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
1232 	 * Alias name of args:
1233 	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
1234 	 * Type of args:
1235 	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
1236 	 */
1237 	struct trace_fprobe *tf __free(free_trace_fprobe) = NULL;
1238 	const char *event = NULL, *group = FPROBE_EVENT_SYSTEM;
1239 	struct module *mod __free(module_put) = NULL;
1240 	const char **new_argv __free(kfree) = NULL;
1241 	char *symbol __free(kfree) = NULL;
1242 	char *ebuf __free(kfree) = NULL;
1243 	char *gbuf __free(kfree) = NULL;
1244 	char *sbuf __free(kfree) = NULL;
1245 	char *abuf __free(kfree) = NULL;
1246 	char *dbuf __free(kfree) = NULL;
1247 	int i, new_argc = 0, ret = 0;
1248 	bool is_tracepoint = false;
1249 	bool is_return = false;
1250 
1251 	if ((argv[0][0] != 'f' && argv[0][0] != 't') || argc < 2)
1252 		return -ECANCELED;
1253 
1254 	if (argv[0][0] == 't') {
1255 		is_tracepoint = true;
1256 		group = TRACEPOINT_EVENT_SYSTEM;
1257 	}
1258 
1259 	if (argv[0][1] != '\0') {
1260 		if (argv[0][1] != ':') {
1261 			trace_probe_log_set_index(0);
1262 			trace_probe_log_err(1, BAD_MAXACT);
1263 			return -EINVAL;
1264 		}
1265 		event = &argv[0][2];
1266 	}
1267 
1268 	trace_probe_log_set_index(1);
1269 
1270 	/* a symbol(or tracepoint) must be specified */
1271 	ret = parse_symbol_and_return(argc, argv, &symbol, &is_return, is_tracepoint);
1272 	if (ret < 0)
1273 		return -EINVAL;
1274 
1275 	trace_probe_log_set_index(0);
1276 	if (event) {
1277 		gbuf = kmalloc(MAX_EVENT_NAME_LEN, GFP_KERNEL);
1278 		if (!gbuf)
1279 			return -ENOMEM;
1280 		ret = traceprobe_parse_event_name(&event, &group, gbuf,
1281 						  event - argv[0]);
1282 		if (ret)
1283 			return -EINVAL;
1284 	}
1285 
1286 	if (!event) {
1287 		ebuf = kmalloc(MAX_EVENT_NAME_LEN, GFP_KERNEL);
1288 		if (!ebuf)
1289 			return -ENOMEM;
1290 		/* Make a new event name */
1291 		if (is_tracepoint)
1292 			snprintf(ebuf, MAX_EVENT_NAME_LEN, "%s%s",
1293 				 isdigit(*symbol) ? "_" : "", symbol);
1294 		else
1295 			snprintf(ebuf, MAX_EVENT_NAME_LEN, "%s__%s", symbol,
1296 				 is_return ? "exit" : "entry");
1297 		sanitize_event_name(ebuf);
1298 		event = ebuf;
1299 	}
1300 
1301 	if (is_return)
1302 		ctx->flags |= TPARG_FL_RETURN;
1303 	else
1304 		ctx->flags |= TPARG_FL_FENTRY;
1305 
1306 	ctx->funcname = NULL;
1307 	if (is_tracepoint) {
1308 		/* Get tracepoint and lock its module until the end of the registration. */
1309 		struct tracepoint *tpoint;
1310 
1311 		ctx->flags |= TPARG_FL_TPOINT;
1312 		mod = NULL;
1313 		tpoint = find_tracepoint(symbol, &mod);
1314 		if (tpoint) {
1315 			sbuf = kmalloc(KSYM_NAME_LEN, GFP_KERNEL);
1316 			if (!sbuf)
1317 				return -ENOMEM;
1318 			ctx->funcname = kallsyms_lookup((unsigned long)tpoint->probestub,
1319 							NULL, NULL, NULL, sbuf);
1320 		}
1321 	}
1322 	if (!ctx->funcname)
1323 		ctx->funcname = symbol;
1324 
1325 	abuf = kmalloc(MAX_BTF_ARGS_LEN, GFP_KERNEL);
1326 	if (!abuf)
1327 		return -ENOMEM;
1328 	argc -= 2; argv += 2;
1329 	new_argv = traceprobe_expand_meta_args(argc, argv, &new_argc,
1330 					       abuf, MAX_BTF_ARGS_LEN, ctx);
1331 	if (IS_ERR(new_argv))
1332 		return PTR_ERR(new_argv);
1333 	if (new_argv) {
1334 		argc = new_argc;
1335 		argv = new_argv;
1336 	}
1337 	if (argc > MAX_TRACE_ARGS) {
1338 		trace_probe_log_set_index(2);
1339 		trace_probe_log_err(0, TOO_MANY_ARGS);
1340 		return -E2BIG;
1341 	}
1342 
1343 	ret = traceprobe_expand_dentry_args(argc, argv, &dbuf);
1344 	if (ret)
1345 		return ret;
1346 
1347 	/* setup a probe */
1348 	tf = alloc_trace_fprobe(group, event, symbol, argc, is_return, is_tracepoint);
1349 	if (IS_ERR(tf)) {
1350 		ret = PTR_ERR(tf);
1351 		/* This must return -ENOMEM, else there is a bug */
1352 		WARN_ON_ONCE(ret != -ENOMEM);
1353 		return ret;
1354 	}
1355 
1356 	/* parse arguments */
1357 	for (i = 0; i < argc; i++) {
1358 		trace_probe_log_set_index(i + 2);
1359 		ctx->offset = 0;
1360 		ret = traceprobe_parse_probe_arg(&tf->tp, i, argv[i], ctx);
1361 		if (ret)
1362 			return ret;	/* This can be -ENOMEM */
1363 	}
1364 
1365 	if (is_return && tf->tp.entry_arg) {
1366 		tf->fp.entry_handler = trace_fprobe_entry_handler;
1367 		tf->fp.entry_data_size = traceprobe_get_entry_data_size(&tf->tp);
1368 		if (ALIGN(tf->fp.entry_data_size, sizeof(long)) > MAX_FPROBE_DATA_SIZE) {
1369 			trace_probe_log_set_index(2);
1370 			trace_probe_log_err(0, TOO_MANY_EARGS);
1371 			return -E2BIG;
1372 		}
1373 	}
1374 
1375 	ret = traceprobe_set_print_fmt(&tf->tp,
1376 			is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL);
1377 	if (ret < 0)
1378 		return ret;
1379 
1380 	ret = register_trace_fprobe_event(tf);
1381 	if (ret) {
1382 		trace_probe_log_set_index(1);
1383 		if (ret == -EILSEQ)
1384 			trace_probe_log_err(0, BAD_INSN_BNDRY);
1385 		else if (ret == -ENOENT)
1386 			trace_probe_log_err(0, BAD_PROBE_ADDR);
1387 		else if (ret != -ENOMEM && ret != -EEXIST)
1388 			trace_probe_log_err(0, FAIL_REG_PROBE);
1389 		return -EINVAL;
1390 	}
1391 
1392 	/* 'tf' is successfully registered. To avoid freeing, assign NULL. */
1393 	tf = NULL;
1394 
1395 	return 0;
1396 }
1397 
trace_fprobe_create_cb(int argc,const char * argv[])1398 static int trace_fprobe_create_cb(int argc, const char *argv[])
1399 {
1400 	struct traceprobe_parse_context *ctx __free(traceprobe_parse_context) = NULL;
1401 	int ret;
1402 
1403 	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1404 	if (!ctx)
1405 		return -ENOMEM;
1406 
1407 	ctx->flags = TPARG_FL_KERNEL | TPARG_FL_FPROBE;
1408 
1409 	trace_probe_log_init("trace_fprobe", argc, argv);
1410 	ret = trace_fprobe_create_internal(argc, argv, ctx);
1411 	trace_probe_log_clear();
1412 	return ret;
1413 }
1414 
trace_fprobe_create(const char * raw_command)1415 static int trace_fprobe_create(const char *raw_command)
1416 {
1417 	return trace_probe_create(raw_command, trace_fprobe_create_cb);
1418 }
1419 
trace_fprobe_release(struct dyn_event * ev)1420 static int trace_fprobe_release(struct dyn_event *ev)
1421 {
1422 	struct trace_fprobe *tf = to_trace_fprobe(ev);
1423 	int ret = unregister_trace_fprobe(tf);
1424 
1425 	if (!ret)
1426 		free_trace_fprobe(tf);
1427 	return ret;
1428 }
1429 
trace_fprobe_show(struct seq_file * m,struct dyn_event * ev)1430 static int trace_fprobe_show(struct seq_file *m, struct dyn_event *ev)
1431 {
1432 	struct trace_fprobe *tf = to_trace_fprobe(ev);
1433 	int i;
1434 
1435 	if (trace_fprobe_is_tracepoint(tf))
1436 		seq_putc(m, 't');
1437 	else
1438 		seq_putc(m, 'f');
1439 	seq_printf(m, ":%s/%s", trace_probe_group_name(&tf->tp),
1440 				trace_probe_name(&tf->tp));
1441 
1442 	seq_printf(m, " %s%s", trace_fprobe_symbol(tf),
1443 			       trace_fprobe_is_return(tf) ? "%return" : "");
1444 
1445 	for (i = 0; i < tf->tp.nr_args; i++)
1446 		seq_printf(m, " %s=%s", tf->tp.args[i].name, tf->tp.args[i].comm);
1447 	seq_putc(m, '\n');
1448 
1449 	return 0;
1450 }
1451 
1452 /*
1453  * Enable trace_probe
1454  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
1455  */
enable_trace_fprobe(struct trace_event_call * call,struct trace_event_file * file)1456 static int enable_trace_fprobe(struct trace_event_call *call,
1457 			       struct trace_event_file *file)
1458 {
1459 	struct trace_probe *tp;
1460 	struct trace_fprobe *tf;
1461 	bool enabled;
1462 	int ret = 0;
1463 
1464 	tp = trace_probe_primary_from_call(call);
1465 	if (WARN_ON_ONCE(!tp))
1466 		return -ENODEV;
1467 	enabled = trace_probe_is_enabled(tp);
1468 
1469 	/* This also changes "enabled" state */
1470 	if (file) {
1471 		ret = trace_probe_add_file(tp, file);
1472 		if (ret)
1473 			return ret;
1474 	} else
1475 		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1476 
1477 	if (!enabled) {
1478 		list_for_each_entry(tf, trace_probe_probe_list(tp), tp.list) {
1479 			ret = __register_trace_fprobe(tf);
1480 			if (ret < 0)
1481 				return ret;
1482 		}
1483 	}
1484 
1485 	return 0;
1486 }
1487 
1488 /*
1489  * Disable trace_probe
1490  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
1491  */
disable_trace_fprobe(struct trace_event_call * call,struct trace_event_file * file)1492 static int disable_trace_fprobe(struct trace_event_call *call,
1493 				struct trace_event_file *file)
1494 {
1495 	struct trace_fprobe *tf;
1496 	struct trace_probe *tp;
1497 
1498 	tp = trace_probe_primary_from_call(call);
1499 	if (WARN_ON_ONCE(!tp))
1500 		return -ENODEV;
1501 
1502 	if (file) {
1503 		if (!trace_probe_get_file_link(tp, file))
1504 			return -ENOENT;
1505 		if (!trace_probe_has_single_file(tp))
1506 			goto out;
1507 		trace_probe_clear_flag(tp, TP_FLAG_TRACE);
1508 	} else
1509 		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1510 
1511 	if (!trace_probe_is_enabled(tp)) {
1512 		list_for_each_entry(tf, trace_probe_probe_list(tp), tp.list) {
1513 			unregister_fprobe(&tf->fp);
1514 		}
1515 	}
1516 
1517  out:
1518 	if (file)
1519 		/*
1520 		 * Synchronization is done in below function. For perf event,
1521 		 * file == NULL and perf_trace_event_unreg() calls
1522 		 * tracepoint_synchronize_unregister() to ensure synchronize
1523 		 * event. We don't need to care about it.
1524 		 */
1525 		trace_probe_remove_file(tp, file);
1526 
1527 	return 0;
1528 }
1529 
1530 /*
1531  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1532  */
fprobe_register(struct trace_event_call * event,enum trace_reg type,void * data)1533 static int fprobe_register(struct trace_event_call *event,
1534 			   enum trace_reg type, void *data)
1535 {
1536 	struct trace_event_file *file = data;
1537 
1538 	switch (type) {
1539 	case TRACE_REG_REGISTER:
1540 		return enable_trace_fprobe(event, file);
1541 	case TRACE_REG_UNREGISTER:
1542 		return disable_trace_fprobe(event, file);
1543 
1544 #ifdef CONFIG_PERF_EVENTS
1545 	case TRACE_REG_PERF_REGISTER:
1546 		return enable_trace_fprobe(event, NULL);
1547 	case TRACE_REG_PERF_UNREGISTER:
1548 		return disable_trace_fprobe(event, NULL);
1549 	case TRACE_REG_PERF_OPEN:
1550 	case TRACE_REG_PERF_CLOSE:
1551 	case TRACE_REG_PERF_ADD:
1552 	case TRACE_REG_PERF_DEL:
1553 		return 0;
1554 #endif
1555 	}
1556 	return 0;
1557 }
1558 
1559 /*
1560  * Register dynevent at core_initcall. This allows kernel to setup fprobe
1561  * events in postcore_initcall without tracefs.
1562  */
init_fprobe_trace_early(void)1563 static __init int init_fprobe_trace_early(void)
1564 {
1565 	int ret;
1566 
1567 	ret = dyn_event_register(&trace_fprobe_ops);
1568 	if (ret)
1569 		return ret;
1570 
1571 #ifdef CONFIG_MODULES
1572 	ret = register_tracepoint_module_notifier(&tracepoint_module_nb);
1573 	if (ret)
1574 		return ret;
1575 	ret = register_module_notifier(&tprobe_event_module_nb);
1576 	if (ret)
1577 		return ret;
1578 #endif
1579 
1580 	return 0;
1581 }
1582 core_initcall(init_fprobe_trace_early);
1583